research-explorer

Research Explorer

Overview

Research-topic exploration SKILL. Takes a broad direction, performs multi-dimensional web research with the agent's own WebSearch / WebFetch tools, and produces three structured Markdown deliverables. Single stage, full quality from the start. No Python runtime, no LLM SDK.

When to Use

• User says "I want to research X" without a specific topic.
• User wants to know "what are the hot topics in X".
• User needs help narrowing a broad field into 5–10 candidate topics.
• User asks for "research landscape overview".

When NOT to Use

• User already has a specific research question → use literature-survey or paper-writer.
• User wants a quick fact-check → use WebSearch directly.

Workflow

Step 1 — Understand the direction

Confirm with the user:

• Direction — the broad area of interest (e.g., "federated learning", "NLP for healthcare").
• Constraints — theory vs. applied, specific methods, target venue, compute budget, time horizon.
• Language — default English in conversation; reports in English unless the user requests otherwise.

Step 2 — Set up the run directory

DIRECTION="<direction>"
SLUG=$(python3 -c "import re,hashlib,sys; t=sys.argv[1]; n=re.sub(r'[\\s_]+','-',re.sub(r'[^\\w\\s-]','',t.lower().strip())).strip('-')[:40].rstrip('-'); h=hashlib.sha1(t.encode()).hexdigest()[:8]; print(f'{n}-{h}')" "$DIRECTION")
TS=$(date +%Y-%m-%d_%H%M%S)
RUN=output/research-explorer/$SLUG/$TS

mkdir -p "$RUN"
ln -sfn "$TS" "output/research-explorer/$SLUG/latest"

In commands below $RUN = output/research-explorer/<slug>/latest.

Step 3 — Multi-dimensional exploration

Run WebSearch across the following dimensions (one query per dimension, more if returns are thin):

1. Hot topics — " 2024 2025 hot topics" / "recent advances".
2. Open problems — " open problems" / "challenges".
3. Surveys — " survey 2024" / " review".
4. Benchmarks — " benchmark" / " evaluation dataset".
5. Applications — " applications" / " industry use cases".
6. Cross-field — " + " (pick 1–2 adjacent fields).
7. Recent breakthroughs — papers from the last 6–12 months at top venues.